A couple years ago, Annika Enejder confronted confusing results from her studies of lipid storage in the roundworm Caenorhabditis elegans. Fluorescence microscopy images very clearly showed a decrease in signal from lipid droplets when the worms were treated with statins, a widely prescribed class of cholesterol-lowering drugs. But simultaneous experiments using another type of microscopy, which visualizes lipids directly, showed no such thing. In fact, coherent anti-Stokes Raman scattering (CARS) microscopy identified lipid droplets where they could not be observed with fluorescence techniques.

What had happened was that worms fed with the commonly used fluorescent dye Nile red processed it as a toxin: instead of being localized to lipid droplets, the dye was sequestered in intestinal lysozome-like granules interspersed among lipid droplets. In fact, the dye can be misleading in other ways: statins themselves seem to affect either its staining or fluorescence. “There are so many artifacts that you have to take into account when using fluorophores,” says Enejder, a scientist at the Chalmers University of Technology in Göteborg, Sweden.

No one can deny the power of fluorescent probes and molecular stains to see the inner workings of cells, but such labels have considerable drawbacks. Delivering labels can be a problem, particularly for whole organisms. Some labels work only on dead cells; others damage cells or perturb the very processes they are intended to study. Label-free microscopy offers a way to investigate living cells while eliminating a slew of possible artifacts. Though some techniques rely on endogenous fluorophores (Box 1), most eschew fluorescence all together, along with the well-known problem of photobleaching. Instead of detecting photons emitted from excited fluorophores, these alternate techniques detect subtle changes in light as it is absorbed or altered by biological samples, relying on the nonlinear optical phenomena observed when high-intensity light moves through matter. In essence, pulses of laser light can be used to 'see' chemical composition: the C-H bonds of lipids, the amide bonds of proteins, the reduced or oxidized states of certain biomolecules, the regularly repeating units of microtubules or collagen.

Harvard University's Brian Saar, Gary Holtom and Sunney Xie (left to right) are developing new techniques and applications of nonlinear microscopy. Credit: Udo Loster

Of course such techniques have their own limitations: whereas fluorescence labeling can often allow discrimination of single molecules, label-free techniques are less sensitive and specific. All but the most common substituents tend to be hidden in signals generated from a few abundant species. “What is nice is that you don't need any labeling; you can just start imaging,” explains Kees Jalink, a biophysicist at the Netherlands Cancer Institute, “but what is not so nice is that the signal is weak, you need a lot of power to irradiate a cell and may only get the coarsest of details.”

Many modes of nonlinearity

In addition to CARS, some of the other techniques that can be done without adding labels include two-photon absorption, second-harmonic generation (SHG) and stimulated Raman scattering, each with its own detection capabilities and requirements (Table 1). The spread of such technologies to biologists, however, does not occur at light speed. Expensive lasers must be coupled to microscopes; short pulses of light must be precisely aimed, coordinated and shaped; and detection devices must be optimized to home in on signals and discard background. “It's not easy to do the assembly without a lot of expertise; these instruments are pretty finicky,” says Robin F.B. Turner, who holds positions in the chemistry and engineering departments at The University of British Columbia. And just setting up the equipment is not enough. “You have to make adjustments on a day-to-day basis,” adds Turner.

Table 1 Selected label-free nonlinear optical techniques
Full size table

Turner says he has good reason for tackling these techniques: he wants to know how stem cells' composition changes as they differentiate into other types of cells, and there is no better way to answer his questions. “The reason we use Raman and CARS is because we can do that nondestructively,” he says. Other measurements destroy cells, creating isolated snapshots of different cells at given points in time; such data are not particularly useful for heterogeneous cultures containing spontaneously differentiating cells, Turner says, adding, “we want to follow cells as they grow.”

Such advantages apply to organisms as well. Harvard University's Gary Ruvkun, for example, is conducting RNA interference screens in C. elegans to probe thousands of genes for their role in lipidogenesis and using a microscopy technique called stimulated Raman scattering (SRS) to monitor the results.

Ruvkun's collaborator is Sunney Xie, also of Harvard University. About ten years ago, Xie made a big splash by describing CARS microscopy, a technique that boosts signals from a phenomenon known as spontaneous Raman scattering, in which chemical bonds in a sample shift the wavelength of light passing through it1. Early implementations of Raman microscopy required powerful lasers and sometimes required exposure times as a long as a day. Xie and colleagues showed CARS could work in living cells. The feeble Raman signal coming from cells could be amplified by using two laser beams whose frequency difference is tuned to the vibrational frequency of the molecular bonds to be imaged. “It provides orders of magnitude more sensitivity than spontaneous Raman,” says Xie. But CARS has drawbacks. It typically focuses on just one part of the broad Raman spectrum at a time, restricting the amount of information collected; it also produces a high background signal. In practical terms, these limitations mean that most applications of CARS involve the detection of lipids, whose high concentrations of carbon-hydrogen bonds create a strong characteristic signal.

A lipid-rich sebaceous gland wraps around a protein rich-hair. Image taken with stimulated Raman scattering, showing lipids in green and proteins in blue. Credit: Christian Freudiger

Xie's enthusiasm has shifted to SRS, which Xie developed with lab members Wei Min and Christian Freudiger and reported late in 2008 (ref. 2). “In CARS, the peak is shifted,” explains Xie; “that means we cannot use the wealth of Raman literature to assign chemical species.” In contrast, he says, SRS eliminates the background signal of CARS by using an incredibly fast and precise modulation of the lasers. Not only does this produce the same spectrum as that obtained in classical Raman spectroscopy, but its signal is orders of magnitude stronger and requires much less time to detect than non-amplified Raman signals. Even better, Xie says, the signal from SRS rises linearly with the number of vibrating bonds, making the technique quantitative. Applications include real-time observations: the absorption of retinoic acid into the skin, for example, has applications for drugs and cosmetics. The ability to monitor how either acids or enzymes remove lignin from plant cell walls could be used to produce biofuels more efficiently.

These proof-of-principle studies have grown from Xie's collaborations with researchers at Pfizer and Harvard University. Xie goes so far as to predict that SRS will supersede CARS, but other researchers are not so sure. SRS requires mixing and understanding signals from multiple light sources and overlapping spectra can be hard to deconvolute. Turner says he has tried using SRS to look at nucleic acids in solution and decided to stick with the techniques already used in his lab. With those, he can distinguish RNA from DNA in cells. Whereas a Raman microscope might be slow, he says, “it would be quite an undertaking to apply SRS and use it to advance our knowledge as fast as plain-Jane Raman.”

Something to buy, something to share

Xie predicts SRS adoption will spread once it has been built into a commercially available system, an advance that he believes will happen later this year; both Zeiss and Leica are reported to have licensed the technology last year. But, as an example from fluorescence microscopy shows, the spread of a technology can be slow. The first commercial multiphoton microscope was introduced in 1996; a survey in 2003 found that 66% of biological research studies using multiphoton microscopy still used custom-built systems3. Today, commercial multiphoton microscopes are commonplace.

In October 2009 Olympus announced plans to offer a femtoCARS module that can be added to multiphoton microscopes, ten years after Xie's publication. In January of 2010, Newport Corporation introduced its wavelength-extension unit that can be attached to a laser and multiphoton microscope to support CARS, SHG and other forms of imaging. Leica will reportedly roll out its own product later this year. Olympus product manager Yiwei (Kevin) Jia says he was helping research groups get started with CARS even before the femtoCARS module was announced; this add-on, set up to detect lipids, will make getting started much easier. If the spread of commercial products for CARS imitates that of multiphoton microscopy, sales will take off in a few years, he says. Right now, though, most researchers doing CARS microscopy are doing so out of physics labs using systems they built themselves.

Microscope and optics set up for spontaneous Raman imaging. Other techniques require a more complex setup. CCD, charge-coupled device. Credit: Robin Turner

Still, these researchers are collaborating with biologists. At Purdue University, biomedical engineering professor Ji-Xen Cheng uses CARS to quickly find lipid bodies within a cell and then, using the same laser source, switches to confocal Raman for a finer analysis of chemical composition within that area. Recent work on human prostate tumor cells showed that regions previously considered to be composed of fat are, surprisingly, oxidized fatty acids. The next step is to determine whether such fatty acids can be used as an indicator of how aggressive the cancer is likely to be. In other work, Cheng has developed a platform to collect signals simultaneously from CARS, which sees lipid, and from a technique called sum frequency generation, which can visualize certain protein fibers. This allowed Cheng and his collaborators to study how lipid-rich immune cells embed themselves in the collagen matrix found in the walls of blood vessels—observations that should lead to insights about how plaques form in atherosclerosis. Separately, Cheng and colleagues monitored the myelin sheath of neurons in mouse models of multiple sclerosis and pinpointed exactly where on the axons damage occurs. “Before,” Cheng says, “there was no way to visualize the myelin at this single-cell resolution in live tissues.”

Label-free CARS imaging of the same axon in live spinal tissue shows that damage to myelin can be induced by glutamate excitotoxicity. Scale bar, 10 μm. Credit: Ji-Xin Cheng

Myelin, with its densely packed lipids, is particularly suited for CARS microscopy, says Jalink. Other applications of label-free microscopy face a high bar. Researchers who work regularly with lasers could probably figure out a way to adopt such techniques, he says, adding, “technically, it's completely doable, but what is the reason for going through this somewhat complicated, expensive process if I can get the information another way?”

As techniques become established, researchers can apply them to new uses. Rafael Yuste at Columbia University is using optics to measure the voltage of neurons. SHG imaging relies on a hyperscattering of light caused by a very regular arrangement of molecules with a strong induced dipole moment or a requisite type of charge distribution. Yuste is particularly interested in such molecules in cell membranes of neurons, across which there is an electric field. As the second harmonic signal is directly proportional to the strength of the electric field, it automatically reports the voltage.

The problem is that few appropriate molecules have been identified that can work well for this purpose. To get to the cutting edge, Yuste says, “you need to scan the whole spectrum carefully, looking for potential second harmonic endogenous chromophores.” The lack of such resources, he says, comes from the fact that the development of such techniques comes from intersections of disciplines, requiring work on the periphery of researchers' expertise, often with less funding and support in researchers' own departments.

A cell-seeded scaffold shown with overlaid images obtained with CARS (muscle cells in orange) and SHG (cellulose and collagen fibers in blue) microscopy. Volume shown is 100 × 100 × 16 μm. Credit: Annika Enejder

Indeed, researchers like Enejder believe that moving across disciplines can help researchers uncover a wealth of problems that can be addressed only with label-free nonlinear microscopy. Though trained as a physicist, Enejder transferred to her university's biology department where it is easier to learn what researchers are having trouble visualizing and how nonlinear optics might help. Researchers who are closeted in physics departments can continue technical developments, she says, but they may not know what biologists hope to see: “I don't have any problem with that at all. I see applications everywhere.”

And while such communication is certainly important so, too, is willingness to attempt experiments that defy physicists' previous experience. In one bioengineering project aimed at producing elastic blood vessels, Enejder and colleagues wanted to monitor the growth of muscle cells seeded into a cellulose matrix. Along with CARS, Enejder and colleagues used SHG to watch the seeded cells and found to their delight that they could monitor how the seeded cells made contact with the cellulose network and began to lay down collagen fibers, a crucial ability for defining optimal parameters for tissue engineering. Though plant-produced cellulose fibers in paper are invisible to SHG, those secreted by bacteria have the kind of regular pattern that does generate an SHG signal, Enejder explains, adding, “you can't rely on different reports of what should be visualized. It is far more reliable to just try it yourself.” See Table 2

Table 2 Suppliers Guide: Companies offering nonlinear or label-free imaging equipment
Full size table